4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Author: Eric Mei <ericm@clusterfs.com>
37 #define DEBUG_SUBSYSTEM S_SEC
39 #include <linux/user_namespace.h>
40 #ifdef HAVE_UIDGID_HEADER
41 # include <linux/uidgid.h>
43 #include <linux/crypto.h>
44 #include <linux/key.h>
46 #include <libcfs/libcfs.h>
48 #include <obd_class.h>
49 #include <obd_support.h>
50 #include <lustre_net.h>
51 #include <lustre_import.h>
52 #include <lustre_dlm.h>
53 #include <lustre_sec.h>
55 #include "ptlrpc_internal.h"
57 static int send_sepol;
58 module_param(send_sepol, int, 0644);
59 MODULE_PARM_DESC(send_sepol, "Client sends SELinux policy status");
65 static rwlock_t policy_lock;
66 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
70 int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
72 __u16 number = policy->sp_policy;
74 LASSERT(policy->sp_name);
75 LASSERT(policy->sp_cops);
76 LASSERT(policy->sp_sops);
78 if (number >= SPTLRPC_POLICY_MAX)
81 write_lock(&policy_lock);
82 if (unlikely(policies[number])) {
83 write_unlock(&policy_lock);
86 policies[number] = policy;
87 write_unlock(&policy_lock);
89 CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
92 EXPORT_SYMBOL(sptlrpc_register_policy);
94 int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
96 __u16 number = policy->sp_policy;
98 LASSERT(number < SPTLRPC_POLICY_MAX);
100 write_lock(&policy_lock);
101 if (unlikely(policies[number] == NULL)) {
102 write_unlock(&policy_lock);
103 CERROR("%s: already unregistered\n", policy->sp_name);
107 LASSERT(policies[number] == policy);
108 policies[number] = NULL;
109 write_unlock(&policy_lock);
111 CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
114 EXPORT_SYMBOL(sptlrpc_unregister_policy);
117 struct ptlrpc_sec_policy *sptlrpc_wireflavor2policy(__u32 flavor)
119 static DEFINE_MUTEX(load_mutex);
120 static atomic_t loaded = ATOMIC_INIT(0);
121 struct ptlrpc_sec_policy *policy;
122 __u16 number = SPTLRPC_FLVR_POLICY(flavor);
125 if (number >= SPTLRPC_POLICY_MAX)
129 read_lock(&policy_lock);
130 policy = policies[number];
131 if (policy && !try_module_get(policy->sp_owner))
134 flag = atomic_read(&loaded);
135 read_unlock(&policy_lock);
137 if (policy != NULL || flag != 0 ||
138 number != SPTLRPC_POLICY_GSS)
141 /* try to load gss module, once */
142 mutex_lock(&load_mutex);
143 if (atomic_read(&loaded) == 0) {
144 if (request_module("ptlrpc_gss") == 0)
146 "module ptlrpc_gss loaded on demand\n");
148 CERROR("Unable to load module ptlrpc_gss\n");
150 atomic_set(&loaded, 1);
152 mutex_unlock(&load_mutex);
158 __u32 sptlrpc_name2flavor_base(const char *name)
160 if (!strcmp(name, "null"))
161 return SPTLRPC_FLVR_NULL;
162 if (!strcmp(name, "plain"))
163 return SPTLRPC_FLVR_PLAIN;
164 if (!strcmp(name, "gssnull"))
165 return SPTLRPC_FLVR_GSSNULL;
166 if (!strcmp(name, "krb5n"))
167 return SPTLRPC_FLVR_KRB5N;
168 if (!strcmp(name, "krb5a"))
169 return SPTLRPC_FLVR_KRB5A;
170 if (!strcmp(name, "krb5i"))
171 return SPTLRPC_FLVR_KRB5I;
172 if (!strcmp(name, "krb5p"))
173 return SPTLRPC_FLVR_KRB5P;
174 if (!strcmp(name, "skn"))
175 return SPTLRPC_FLVR_SKN;
176 if (!strcmp(name, "ska"))
177 return SPTLRPC_FLVR_SKA;
178 if (!strcmp(name, "ski"))
179 return SPTLRPC_FLVR_SKI;
180 if (!strcmp(name, "skpi"))
181 return SPTLRPC_FLVR_SKPI;
183 return SPTLRPC_FLVR_INVALID;
185 EXPORT_SYMBOL(sptlrpc_name2flavor_base);
187 const char *sptlrpc_flavor2name_base(__u32 flvr)
189 __u32 base = SPTLRPC_FLVR_BASE(flvr);
191 if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL))
193 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN))
195 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_GSSNULL))
197 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
199 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
201 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
203 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
205 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKN))
207 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKA))
209 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKI))
211 else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKPI))
214 CERROR("invalid wire flavor 0x%x\n", flvr);
217 EXPORT_SYMBOL(sptlrpc_flavor2name_base);
219 char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
220 char *buf, int bufsize)
222 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
223 snprintf(buf, bufsize, "hash:%s",
224 sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
226 snprintf(buf, bufsize, "%s",
227 sptlrpc_flavor2name_base(sf->sf_rpc));
229 buf[bufsize - 1] = '\0';
232 EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
234 char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
236 snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
239 * currently we don't support customized bulk specification for
240 * flavors other than plain
242 if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
246 sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
247 strncat(buf, bspec, bufsize);
250 buf[bufsize - 1] = '\0';
253 EXPORT_SYMBOL(sptlrpc_flavor2name);
255 char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
259 if (flags & PTLRPC_SEC_FL_REVERSE)
260 strlcat(buf, "reverse,", bufsize);
261 if (flags & PTLRPC_SEC_FL_ROOTONLY)
262 strlcat(buf, "rootonly,", bufsize);
263 if (flags & PTLRPC_SEC_FL_UDESC)
264 strlcat(buf, "udesc,", bufsize);
265 if (flags & PTLRPC_SEC_FL_BULK)
266 strlcat(buf, "bulk,", bufsize);
268 strlcat(buf, "-,", bufsize);
272 EXPORT_SYMBOL(sptlrpc_secflags2str);
275 * client context APIs
279 struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
281 struct vfs_cred vcred;
282 int create = 1, remove_dead = 1;
285 LASSERT(sec->ps_policy->sp_cops->lookup_ctx);
287 if (sec->ps_flvr.sf_flags & (PTLRPC_SEC_FL_REVERSE |
288 PTLRPC_SEC_FL_ROOTONLY)) {
291 if (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE) {
296 vcred.vc_uid = from_kuid(&init_user_ns, current_uid());
297 vcred.vc_gid = from_kgid(&init_user_ns, current_gid());
300 return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred, create,
304 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
306 atomic_inc(&ctx->cc_refcount);
309 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
311 void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
313 struct ptlrpc_sec *sec = ctx->cc_sec;
316 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
318 if (!atomic_dec_and_test(&ctx->cc_refcount))
321 sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
323 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
326 * Expire the client context immediately.
328 * \pre Caller must hold at least 1 reference on the \a ctx.
330 void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
332 LASSERT(ctx->cc_ops->die);
333 ctx->cc_ops->die(ctx, 0);
335 EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
338 * To wake up the threads who are waiting for this client context. Called
339 * after some status change happened on \a ctx.
341 void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
343 struct ptlrpc_request *req, *next;
345 spin_lock(&ctx->cc_lock);
346 list_for_each_entry_safe(req, next, &ctx->cc_req_list,
348 list_del_init(&req->rq_ctx_chain);
349 ptlrpc_client_wake_req(req);
351 spin_unlock(&ctx->cc_lock);
353 EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
355 int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
357 LASSERT(ctx->cc_ops);
359 if (ctx->cc_ops->display == NULL)
362 return ctx->cc_ops->display(ctx, buf, bufsize);
365 static int import_sec_check_expire(struct obd_import *imp)
369 write_lock(&imp->imp_sec_lock);
370 if (imp->imp_sec_expire &&
371 imp->imp_sec_expire < ktime_get_real_seconds()) {
373 imp->imp_sec_expire = 0;
375 write_unlock(&imp->imp_sec_lock);
380 CDEBUG(D_SEC, "found delayed sec adapt expired, do it now\n");
381 return sptlrpc_import_sec_adapt(imp, NULL, NULL);
385 * Get and validate the client side ptlrpc security facilities from
386 * \a imp. There is a race condition on client reconnect when the import is
387 * being destroyed while there are outstanding client bound requests. In
388 * this case do not output any error messages if import secuity is not
391 * \param[in] imp obd import associated with client
392 * \param[out] sec client side ptlrpc security
394 * \retval 0 if security retrieved successfully
395 * \retval -ve errno if there was a problem
397 static int import_sec_validate_get(struct obd_import *imp,
398 struct ptlrpc_sec **sec)
402 if (unlikely(imp->imp_sec_expire)) {
403 rc = import_sec_check_expire(imp);
408 *sec = sptlrpc_import_sec_ref(imp);
410 CERROR("import %p (%s) with no sec\n",
411 imp, ptlrpc_import_state_name(imp->imp_state));
415 if (unlikely((*sec)->ps_dying)) {
416 CERROR("attempt to use dying sec %p\n", sec);
417 sptlrpc_sec_put(*sec);
425 * Given a \a req, find or allocate an appropriate context for it.
426 * \pre req->rq_cli_ctx == NULL.
428 * \retval 0 succeed, and req->rq_cli_ctx is set.
429 * \retval -ev error number, and req->rq_cli_ctx == NULL.
431 int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
433 struct obd_import *imp = req->rq_import;
434 struct ptlrpc_sec *sec;
439 LASSERT(!req->rq_cli_ctx);
442 rc = import_sec_validate_get(imp, &sec);
446 req->rq_cli_ctx = get_my_ctx(sec);
448 sptlrpc_sec_put(sec);
450 if (!req->rq_cli_ctx) {
451 CERROR("req %p: fail to get context\n", req);
452 RETURN(-ECONNREFUSED);
459 * Drop the context for \a req.
460 * \pre req->rq_cli_ctx != NULL.
461 * \post req->rq_cli_ctx == NULL.
463 * If \a sync == 0, this function should return quickly without sleep;
464 * otherwise it might trigger and wait for the whole process of sending
465 * an context-destroying rpc to server.
467 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
472 LASSERT(req->rq_cli_ctx);
475 * request might be asked to release earlier while still
476 * in the context waiting list.
478 if (!list_empty(&req->rq_ctx_chain)) {
479 spin_lock(&req->rq_cli_ctx->cc_lock);
480 list_del_init(&req->rq_ctx_chain);
481 spin_unlock(&req->rq_cli_ctx->cc_lock);
484 sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
485 req->rq_cli_ctx = NULL;
490 int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
491 struct ptlrpc_cli_ctx *oldctx,
492 struct ptlrpc_cli_ctx *newctx)
494 struct sptlrpc_flavor old_flvr;
495 char *reqmsg = NULL; /* to workaround old gcc */
499 LASSERT(req->rq_reqmsg);
500 LASSERT(req->rq_reqlen);
501 LASSERT(req->rq_replen);
504 "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), switch sec %p(%s) -> %p(%s)\n",
505 req, oldctx, oldctx->cc_vcred.vc_uid,
506 sec2target_str(oldctx->cc_sec), newctx, newctx->cc_vcred.vc_uid,
507 sec2target_str(newctx->cc_sec), oldctx->cc_sec,
508 oldctx->cc_sec->ps_policy->sp_name, newctx->cc_sec,
509 newctx->cc_sec->ps_policy->sp_name);
512 old_flvr = req->rq_flvr;
514 /* save request message */
515 reqmsg_size = req->rq_reqlen;
516 if (reqmsg_size != 0) {
517 OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
520 memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
523 /* release old req/rep buf */
524 req->rq_cli_ctx = oldctx;
525 sptlrpc_cli_free_reqbuf(req);
526 sptlrpc_cli_free_repbuf(req);
527 req->rq_cli_ctx = newctx;
529 /* recalculate the flavor */
530 sptlrpc_req_set_flavor(req, 0);
533 * alloc new request buffer
534 * we don't need to alloc reply buffer here, leave it to the
535 * rest procedure of ptlrpc
537 if (reqmsg_size != 0) {
538 rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
540 LASSERT(req->rq_reqmsg);
541 memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
543 CWARN("failed to alloc reqbuf: %d\n", rc);
544 req->rq_flvr = old_flvr;
547 OBD_FREE_LARGE(reqmsg, reqmsg_size);
553 * If current context of \a req is dead somehow, e.g. we just switched flavor
554 * thus marked original contexts dead, we'll find a new context for it. if
555 * no switch is needed, \a req will end up with the same context.
557 * \note a request must have a context, to keep other parts of code happy.
558 * In any case of failure during the switching, we must restore the old one.
560 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
562 struct ptlrpc_cli_ctx *oldctx = req->rq_cli_ctx;
563 struct ptlrpc_cli_ctx *newctx;
570 sptlrpc_cli_ctx_get(oldctx);
571 sptlrpc_req_put_ctx(req, 0);
573 rc = sptlrpc_req_get_ctx(req);
575 LASSERT(!req->rq_cli_ctx);
577 /* restore old ctx */
578 req->rq_cli_ctx = oldctx;
582 newctx = req->rq_cli_ctx;
585 if (unlikely(newctx == oldctx &&
586 test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
588 * still get the old dead ctx, usually means system too busy
591 "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
592 newctx, newctx->cc_flags);
594 set_current_state(TASK_INTERRUPTIBLE);
595 schedule_timeout(cfs_time_seconds(1));
596 } else if (unlikely(test_bit(PTLRPC_CTX_UPTODATE_BIT, &newctx->cc_flags)
599 * new ctx not up to date yet
602 "ctx (%p, fl %lx) doesn't switch, not up to date yet\n",
603 newctx, newctx->cc_flags);
606 * it's possible newctx == oldctx if we're switching
607 * subflavor with the same sec.
609 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
611 /* restore old ctx */
612 sptlrpc_req_put_ctx(req, 0);
613 req->rq_cli_ctx = oldctx;
617 LASSERT(req->rq_cli_ctx == newctx);
620 sptlrpc_cli_ctx_put(oldctx, 1);
623 EXPORT_SYMBOL(sptlrpc_req_replace_dead_ctx);
626 int ctx_check_refresh(struct ptlrpc_cli_ctx *ctx)
628 if (cli_ctx_is_refreshed(ctx))
634 int ctx_refresh_timeout(struct ptlrpc_request *req)
638 /* conn_cnt is needed in expire_one_request */
639 lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
641 rc = ptlrpc_expire_one_request(req, 1);
643 * if we started recovery, we should mark this ctx dead; otherwise
644 * in case of lgssd died nobody would retire this ctx, following
645 * connecting will still find the same ctx thus cause deadlock.
646 * there's an assumption that expire time of the request should be
647 * later than the context refresh expire time.
650 req->rq_cli_ctx->cc_ops->die(req->rq_cli_ctx, 0);
655 void ctx_refresh_interrupt(struct ptlrpc_request *req)
658 spin_lock(&req->rq_lock);
660 spin_unlock(&req->rq_lock);
664 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
666 spin_lock(&ctx->cc_lock);
667 if (!list_empty(&req->rq_ctx_chain))
668 list_del_init(&req->rq_ctx_chain);
669 spin_unlock(&ctx->cc_lock);
673 * To refresh the context of \req, if it's not up-to-date.
676 * - = 0: wait until success or fatal error occur
677 * - > 0: timeout value (in seconds)
679 * The status of the context could be subject to be changed by other threads
680 * at any time. We allow this race, but once we return with 0, the caller will
681 * suppose it's uptodated and keep using it until the owning rpc is done.
683 * \retval 0 only if the context is uptodated.
684 * \retval -ev error number.
686 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
688 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
689 struct ptlrpc_sec *sec;
696 if (req->rq_ctx_init || req->rq_ctx_fini)
700 * during the process a request's context might change type even
701 * (e.g. from gss ctx to null ctx), so each loop we need to re-check
705 rc = import_sec_validate_get(req->rq_import, &sec);
709 if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
710 CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
711 req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
712 req_off_ctx_list(req, ctx);
713 sptlrpc_req_replace_dead_ctx(req);
714 ctx = req->rq_cli_ctx;
716 sptlrpc_sec_put(sec);
718 if (cli_ctx_is_eternal(ctx))
721 if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
722 LASSERT(ctx->cc_ops->refresh);
723 ctx->cc_ops->refresh(ctx);
725 LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
727 LASSERT(ctx->cc_ops->validate);
728 if (ctx->cc_ops->validate(ctx) == 0) {
729 req_off_ctx_list(req, ctx);
733 if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
734 spin_lock(&req->rq_lock);
736 spin_unlock(&req->rq_lock);
737 req_off_ctx_list(req, ctx);
742 * There's a subtle issue for resending RPCs, suppose following
744 * 1. the request was sent to server.
745 * 2. recovery was kicked start, after finished the request was
747 * 3. resend the request.
748 * 4. old reply from server received, we accept and verify the reply.
749 * this has to be success, otherwise the error will be aware
751 * 5. new reply from server received, dropped by LNet.
753 * Note the xid of old & new request is the same. We can't simply
754 * change xid for the resent request because the server replies on
755 * it for reply reconstruction.
757 * Commonly the original context should be uptodate because we
758 * have an expiry nice time; server will keep its context because
759 * we at least hold a ref of old context which prevent context
760 * from destroying RPC being sent. So server still can accept the
761 * request and finish the RPC. But if that's not the case:
762 * 1. If server side context has been trimmed, a NO_CONTEXT will
763 * be returned, gss_cli_ctx_verify/unseal will switch to new
765 * 2. Current context never be refreshed, then we are fine: we
766 * never really send request with old context before.
768 if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
769 unlikely(req->rq_reqmsg) &&
770 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
771 req_off_ctx_list(req, ctx);
775 if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
776 req_off_ctx_list(req, ctx);
778 * don't switch ctx if import was deactivated
780 if (req->rq_import->imp_deactive) {
781 spin_lock(&req->rq_lock);
783 spin_unlock(&req->rq_lock);
787 rc = sptlrpc_req_replace_dead_ctx(req);
789 LASSERT(ctx == req->rq_cli_ctx);
790 CERROR("req %p: failed to replace dead ctx %p: %d\n",
792 spin_lock(&req->rq_lock);
794 spin_unlock(&req->rq_lock);
798 ctx = req->rq_cli_ctx;
803 * Now we're sure this context is during upcall, add myself into
806 spin_lock(&ctx->cc_lock);
807 if (list_empty(&req->rq_ctx_chain))
808 list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
809 spin_unlock(&ctx->cc_lock);
812 RETURN(-EWOULDBLOCK);
814 /* Clear any flags that may be present from previous sends */
815 LASSERT(req->rq_receiving_reply == 0);
816 spin_lock(&req->rq_lock);
818 req->rq_timedout = 0;
821 spin_unlock(&req->rq_lock);
823 if (wait_event_idle_timeout(req->rq_reply_waitq,
824 ctx_check_refresh(ctx),
825 cfs_time_seconds(timeout)) == 0) {
827 if (!ctx_refresh_timeout(req) &&
828 l_wait_event_abortable(req->rq_reply_waitq,
829 ctx_check_refresh(ctx))
832 ctx_refresh_interrupt(req);
837 * following cases could lead us here:
838 * - successfully refreshed;
840 * - timedout, and we don't want recover from the failure;
841 * - timedout, and waked up upon recovery finished;
842 * - someone else mark this ctx dead by force;
843 * - someone invalidate the req and call ptlrpc_client_wake_req(),
844 * e.g. ptlrpc_abort_inflight();
846 if (!cli_ctx_is_refreshed(ctx)) {
847 /* timed out or interruptted */
848 req_off_ctx_list(req, ctx);
858 * Initialize flavor settings for \a req, according to \a opcode.
860 * \note this could be called in two situations:
861 * - new request from ptlrpc_pre_req(), with proper @opcode
862 * - old request which changed ctx in the middle, with @opcode == 0
864 void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
866 struct ptlrpc_sec *sec;
868 LASSERT(req->rq_import);
869 LASSERT(req->rq_cli_ctx);
870 LASSERT(req->rq_cli_ctx->cc_sec);
871 LASSERT(req->rq_bulk_read == 0 || req->rq_bulk_write == 0);
873 /* special security flags according to opcode */
877 case MGS_CONFIG_READ:
879 req->rq_bulk_read = 1;
883 req->rq_bulk_write = 1;
886 req->rq_ctx_init = 1;
889 req->rq_ctx_fini = 1;
892 /* init/fini rpc won't be resend, so can't be here */
893 LASSERT(req->rq_ctx_init == 0);
894 LASSERT(req->rq_ctx_fini == 0);
896 /* cleanup flags, which should be recalculated */
897 req->rq_pack_udesc = 0;
898 req->rq_pack_bulk = 0;
902 sec = req->rq_cli_ctx->cc_sec;
904 spin_lock(&sec->ps_lock);
905 req->rq_flvr = sec->ps_flvr;
906 spin_unlock(&sec->ps_lock);
909 * force SVC_NULL for context initiation rpc, SVC_INTG for context
912 if (unlikely(req->rq_ctx_init))
913 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
914 else if (unlikely(req->rq_ctx_fini))
915 flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
917 /* user descriptor flag, null security can't do it anyway */
918 if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
919 (req->rq_flvr.sf_rpc != SPTLRPC_FLVR_NULL))
920 req->rq_pack_udesc = 1;
922 /* bulk security flag */
923 if ((req->rq_bulk_read || req->rq_bulk_write) &&
924 sptlrpc_flavor_has_bulk(&req->rq_flvr))
925 req->rq_pack_bulk = 1;
928 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
930 if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
933 LASSERT(req->rq_clrbuf);
934 if (req->rq_pool || !req->rq_reqbuf)
937 OBD_FREE(req->rq_reqbuf, req->rq_reqbuf_len);
938 req->rq_reqbuf = NULL;
939 req->rq_reqbuf_len = 0;
943 * Given an import \a imp, check whether current user has a valid context
944 * or not. We may create a new context and try to refresh it, and try
945 * repeatedly try in case of non-fatal errors. Return 0 means success.
947 int sptlrpc_import_check_ctx(struct obd_import *imp)
949 struct ptlrpc_sec *sec;
950 struct ptlrpc_cli_ctx *ctx;
951 struct ptlrpc_request *req = NULL;
958 sec = sptlrpc_import_sec_ref(imp);
959 ctx = get_my_ctx(sec);
960 sptlrpc_sec_put(sec);
965 if (cli_ctx_is_eternal(ctx) ||
966 ctx->cc_ops->validate(ctx) == 0) {
967 sptlrpc_cli_ctx_put(ctx, 1);
971 if (cli_ctx_is_error(ctx)) {
972 sptlrpc_cli_ctx_put(ctx, 1);
976 req = ptlrpc_request_cache_alloc(GFP_NOFS);
980 ptlrpc_cli_req_init(req);
981 atomic_set(&req->rq_refcount, 10000);
983 req->rq_import = imp;
984 req->rq_flvr = sec->ps_flvr;
985 req->rq_cli_ctx = ctx;
987 rc = sptlrpc_req_refresh_ctx(req, 0);
988 LASSERT(list_empty(&req->rq_ctx_chain));
989 sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
990 ptlrpc_request_cache_free(req);
996 * Used by ptlrpc client, to perform the pre-defined security transformation
997 * upon the request message of \a req. After this function called,
998 * req->rq_reqmsg is still accessible as clear text.
1000 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
1002 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1008 LASSERT(ctx->cc_sec);
1009 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1012 * we wrap bulk request here because now we can be sure
1013 * the context is uptodate.
1016 rc = sptlrpc_cli_wrap_bulk(req, req->rq_bulk);
1021 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1022 case SPTLRPC_SVC_NULL:
1023 case SPTLRPC_SVC_AUTH:
1024 case SPTLRPC_SVC_INTG:
1025 LASSERT(ctx->cc_ops->sign);
1026 rc = ctx->cc_ops->sign(ctx, req);
1028 case SPTLRPC_SVC_PRIV:
1029 LASSERT(ctx->cc_ops->seal);
1030 rc = ctx->cc_ops->seal(ctx, req);
1037 LASSERT(req->rq_reqdata_len);
1038 LASSERT(req->rq_reqdata_len % 8 == 0);
1039 LASSERT(req->rq_reqdata_len <= req->rq_reqbuf_len);
1045 static int do_cli_unwrap_reply(struct ptlrpc_request *req)
1047 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1053 LASSERT(ctx->cc_sec);
1054 LASSERT(req->rq_repbuf);
1055 LASSERT(req->rq_repdata);
1056 LASSERT(req->rq_repmsg == NULL);
1058 req->rq_rep_swab_mask = 0;
1060 rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
1063 lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
1067 CERROR("failed unpack reply: x%llu\n", req->rq_xid);
1071 if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
1072 CERROR("replied data length %d too small\n",
1073 req->rq_repdata_len);
1077 if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
1078 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
1079 CERROR("reply policy %u doesn't match request policy %u\n",
1080 SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
1081 SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
1085 switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
1086 case SPTLRPC_SVC_NULL:
1087 case SPTLRPC_SVC_AUTH:
1088 case SPTLRPC_SVC_INTG:
1089 LASSERT(ctx->cc_ops->verify);
1090 rc = ctx->cc_ops->verify(ctx, req);
1092 case SPTLRPC_SVC_PRIV:
1093 LASSERT(ctx->cc_ops->unseal);
1094 rc = ctx->cc_ops->unseal(ctx, req);
1099 LASSERT(rc || req->rq_repmsg || req->rq_resend);
1101 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
1103 req->rq_rep_swab_mask = 0;
1108 * Used by ptlrpc client, to perform security transformation upon the reply
1109 * message of \a req. After return successfully, req->rq_repmsg points to
1110 * the reply message in clear text.
1112 * \pre the reply buffer should have been un-posted from LNet, so nothing is
1115 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
1117 LASSERT(req->rq_repbuf);
1118 LASSERT(req->rq_repdata == NULL);
1119 LASSERT(req->rq_repmsg == NULL);
1120 LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
1122 if (req->rq_reply_off == 0 &&
1123 (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
1124 CERROR("real reply with offset 0\n");
1128 if (req->rq_reply_off % 8 != 0) {
1129 CERROR("reply at odd offset %u\n", req->rq_reply_off);
1133 req->rq_repdata = (struct lustre_msg *)
1134 (req->rq_repbuf + req->rq_reply_off);
1135 req->rq_repdata_len = req->rq_nob_received;
1137 return do_cli_unwrap_reply(req);
1141 * Used by ptlrpc client, to perform security transformation upon the early
1142 * reply message of \a req. We expect the rq_reply_off is 0, and
1143 * rq_nob_received is the early reply size.
1145 * Because the receive buffer might be still posted, the reply data might be
1146 * changed at any time, no matter we're holding rq_lock or not. For this reason
1147 * we allocate a separate ptlrpc_request and reply buffer for early reply
1150 * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
1151 * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
1152 * \a *req_ret to release it.
1153 * \retval -ev error number, and \a req_ret will not be set.
1155 int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
1156 struct ptlrpc_request **req_ret)
1158 struct ptlrpc_request *early_req;
1160 int early_bufsz, early_size;
1165 early_req = ptlrpc_request_cache_alloc(GFP_NOFS);
1166 if (early_req == NULL)
1169 ptlrpc_cli_req_init(early_req);
1171 early_size = req->rq_nob_received;
1172 early_bufsz = size_roundup_power2(early_size);
1173 OBD_ALLOC_LARGE(early_buf, early_bufsz);
1174 if (early_buf == NULL)
1175 GOTO(err_req, rc = -ENOMEM);
1177 /* sanity checkings and copy data out, do it inside spinlock */
1178 spin_lock(&req->rq_lock);
1180 if (req->rq_replied) {
1181 spin_unlock(&req->rq_lock);
1182 GOTO(err_buf, rc = -EALREADY);
1185 LASSERT(req->rq_repbuf);
1186 LASSERT(req->rq_repdata == NULL);
1187 LASSERT(req->rq_repmsg == NULL);
1189 if (req->rq_reply_off != 0) {
1190 CERROR("early reply with offset %u\n", req->rq_reply_off);
1191 spin_unlock(&req->rq_lock);
1192 GOTO(err_buf, rc = -EPROTO);
1195 if (req->rq_nob_received != early_size) {
1196 /* even another early arrived the size should be the same */
1197 CERROR("data size has changed from %u to %u\n",
1198 early_size, req->rq_nob_received);
1199 spin_unlock(&req->rq_lock);
1200 GOTO(err_buf, rc = -EINVAL);
1203 if (req->rq_nob_received < sizeof(struct lustre_msg)) {
1204 CERROR("early reply length %d too small\n",
1205 req->rq_nob_received);
1206 spin_unlock(&req->rq_lock);
1207 GOTO(err_buf, rc = -EALREADY);
1210 memcpy(early_buf, req->rq_repbuf, early_size);
1211 spin_unlock(&req->rq_lock);
1213 early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
1214 early_req->rq_flvr = req->rq_flvr;
1215 early_req->rq_repbuf = early_buf;
1216 early_req->rq_repbuf_len = early_bufsz;
1217 early_req->rq_repdata = (struct lustre_msg *) early_buf;
1218 early_req->rq_repdata_len = early_size;
1219 early_req->rq_early = 1;
1220 early_req->rq_reqmsg = req->rq_reqmsg;
1222 rc = do_cli_unwrap_reply(early_req);
1224 DEBUG_REQ(D_ADAPTTO, early_req,
1225 "unwrap early reply: rc = %d", rc);
1229 LASSERT(early_req->rq_repmsg);
1230 *req_ret = early_req;
1234 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1236 OBD_FREE_LARGE(early_buf, early_bufsz);
1238 ptlrpc_request_cache_free(early_req);
1243 * Used by ptlrpc client, to release a processed early reply \a early_req.
1245 * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
1247 void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
1249 LASSERT(early_req->rq_repbuf);
1250 LASSERT(early_req->rq_repdata);
1251 LASSERT(early_req->rq_repmsg);
1253 sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
1254 OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
1255 ptlrpc_request_cache_free(early_req);
1258 /**************************************************
1260 **************************************************/
1263 * "fixed" sec (e.g. null) use sec_id < 0
1265 static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1);
1267 int sptlrpc_get_next_secid(void)
1269 return atomic_inc_return(&sptlrpc_sec_id);
1271 EXPORT_SYMBOL(sptlrpc_get_next_secid);
1274 * client side high-level security APIs
1277 static int sec_cop_flush_ctx_cache(struct ptlrpc_sec *sec, uid_t uid,
1278 int grace, int force)
1280 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1282 LASSERT(policy->sp_cops);
1283 LASSERT(policy->sp_cops->flush_ctx_cache);
1285 return policy->sp_cops->flush_ctx_cache(sec, uid, grace, force);
1288 static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
1290 struct ptlrpc_sec_policy *policy = sec->ps_policy;
1292 LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
1293 LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
1294 LASSERT(policy->sp_cops->destroy_sec);
1296 CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
1298 policy->sp_cops->destroy_sec(sec);
1299 sptlrpc_policy_put(policy);
1302 void sptlrpc_sec_destroy(struct ptlrpc_sec *sec)
1304 sec_cop_destroy_sec(sec);
1306 EXPORT_SYMBOL(sptlrpc_sec_destroy);
1308 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
1310 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1312 if (sec->ps_policy->sp_cops->kill_sec) {
1313 sec->ps_policy->sp_cops->kill_sec(sec);
1315 sec_cop_flush_ctx_cache(sec, -1, 1, 1);
1319 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
1322 atomic_inc(&sec->ps_refcount);
1326 EXPORT_SYMBOL(sptlrpc_sec_get);
1328 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
1331 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1333 if (atomic_dec_and_test(&sec->ps_refcount)) {
1334 sptlrpc_gc_del_sec(sec);
1335 sec_cop_destroy_sec(sec);
1339 EXPORT_SYMBOL(sptlrpc_sec_put);
1342 * policy module is responsible for taking refrence of import
1345 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
1346 struct ptlrpc_svc_ctx *svc_ctx,
1347 struct sptlrpc_flavor *sf,
1348 enum lustre_sec_part sp)
1350 struct ptlrpc_sec_policy *policy;
1351 struct ptlrpc_sec *sec;
1357 LASSERT(imp->imp_dlm_fake == 1);
1359 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
1360 imp->imp_obd->obd_type->typ_name,
1361 imp->imp_obd->obd_name,
1362 sptlrpc_flavor2name(sf, str, sizeof(str)));
1364 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
1365 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
1367 LASSERT(imp->imp_dlm_fake == 0);
1369 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
1370 imp->imp_obd->obd_type->typ_name,
1371 imp->imp_obd->obd_name,
1372 sptlrpc_flavor2name(sf, str, sizeof(str)));
1374 policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
1376 CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
1381 sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
1383 atomic_inc(&sec->ps_refcount);
1387 if (sec->ps_gc_interval && policy->sp_cops->gc_ctx)
1388 sptlrpc_gc_add_sec(sec);
1390 sptlrpc_policy_put(policy);
1396 struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
1398 struct ptlrpc_sec *sec;
1400 read_lock(&imp->imp_sec_lock);
1401 sec = sptlrpc_sec_get(imp->imp_sec);
1402 read_unlock(&imp->imp_sec_lock);
1406 EXPORT_SYMBOL(sptlrpc_import_sec_ref);
1408 static void sptlrpc_import_sec_install(struct obd_import *imp,
1409 struct ptlrpc_sec *sec)
1411 struct ptlrpc_sec *old_sec;
1413 LASSERT_ATOMIC_POS(&sec->ps_refcount);
1415 write_lock(&imp->imp_sec_lock);
1416 old_sec = imp->imp_sec;
1418 write_unlock(&imp->imp_sec_lock);
1421 sptlrpc_sec_kill(old_sec);
1423 /* balance the ref taken by this import */
1424 sptlrpc_sec_put(old_sec);
1429 int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
1431 return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
1435 void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
1441 * To get an appropriate ptlrpc_sec for the \a imp, according to the current
1442 * configuration. Upon called, imp->imp_sec may or may not be NULL.
1444 * - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
1445 * - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
1447 int sptlrpc_import_sec_adapt(struct obd_import *imp,
1448 struct ptlrpc_svc_ctx *svc_ctx,
1449 struct sptlrpc_flavor *flvr)
1451 struct ptlrpc_connection *conn;
1452 struct sptlrpc_flavor sf;
1453 struct ptlrpc_sec *sec, *newsec;
1454 enum lustre_sec_part sp;
1465 conn = imp->imp_connection;
1467 if (svc_ctx == NULL) {
1468 struct client_obd *cliobd = &imp->imp_obd->u.cli;
1470 * normal import, determine flavor from rule set, except
1471 * for mgc the flavor is predetermined.
1473 if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
1474 sf = cliobd->cl_flvr_mgc;
1476 sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
1478 &cliobd->cl_target_uuid,
1481 sp = imp->imp_obd->u.cli.cl_sp_me;
1483 /* reverse import, determine flavor from incoming reqeust */
1486 if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
1487 sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
1488 PTLRPC_SEC_FL_ROOTONLY;
1490 sp = sptlrpc_target_sec_part(imp->imp_obd);
1493 sec = sptlrpc_import_sec_ref(imp);
1497 if (flavor_equal(&sf, &sec->ps_flvr))
1500 CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
1501 imp->imp_obd->obd_name,
1502 obd_uuid2str(&conn->c_remote_uuid),
1503 sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
1504 sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
1505 } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
1506 SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
1507 CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
1508 imp->imp_obd->obd_name,
1509 obd_uuid2str(&conn->c_remote_uuid),
1510 LNET_NIDNET(conn->c_self),
1511 sptlrpc_flavor2name(&sf, str, sizeof(str)));
1514 newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
1516 sptlrpc_import_sec_install(imp, newsec);
1518 CERROR("import %s->%s: failed to create new sec\n",
1519 imp->imp_obd->obd_name,
1520 obd_uuid2str(&conn->c_remote_uuid));
1525 sptlrpc_sec_put(sec);
1529 void sptlrpc_import_sec_put(struct obd_import *imp)
1532 sptlrpc_sec_kill(imp->imp_sec);
1534 sptlrpc_sec_put(imp->imp_sec);
1535 imp->imp_sec = NULL;
1539 static void import_flush_ctx_common(struct obd_import *imp,
1540 uid_t uid, int grace, int force)
1542 struct ptlrpc_sec *sec;
1547 sec = sptlrpc_import_sec_ref(imp);
1551 sec_cop_flush_ctx_cache(sec, uid, grace, force);
1552 sptlrpc_sec_put(sec);
1555 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
1558 * it's important to use grace mode, see explain in
1559 * sptlrpc_req_refresh_ctx()
1561 import_flush_ctx_common(imp, 0, 1, 1);
1564 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
1566 import_flush_ctx_common(imp, from_kuid(&init_user_ns, current_uid()),
1569 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
1571 void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
1573 import_flush_ctx_common(imp, -1, 1, 1);
1575 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
1578 * Used by ptlrpc client to allocate request buffer of \a req. Upon return
1579 * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
1581 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
1583 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1584 struct ptlrpc_sec_policy *policy;
1588 LASSERT(ctx->cc_sec);
1589 LASSERT(ctx->cc_sec->ps_policy);
1590 LASSERT(req->rq_reqmsg == NULL);
1591 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1593 policy = ctx->cc_sec->ps_policy;
1594 rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
1596 LASSERT(req->rq_reqmsg);
1597 LASSERT(req->rq_reqbuf || req->rq_clrbuf);
1599 /* zeroing preallocated buffer */
1601 memset(req->rq_reqmsg, 0, msgsize);
1608 * Used by ptlrpc client to free request buffer of \a req. After this
1609 * req->rq_reqmsg is set to NULL and should not be accessed anymore.
1611 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
1613 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1614 struct ptlrpc_sec_policy *policy;
1617 LASSERT(ctx->cc_sec);
1618 LASSERT(ctx->cc_sec->ps_policy);
1619 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1621 if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
1624 policy = ctx->cc_sec->ps_policy;
1625 policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
1626 req->rq_reqmsg = NULL;
1630 * NOTE caller must guarantee the buffer size is enough for the enlargement
1632 void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
1633 int segment, int newsize)
1636 int oldsize, oldmsg_size, movesize;
1638 LASSERT(segment < msg->lm_bufcount);
1639 LASSERT(msg->lm_buflens[segment] <= newsize);
1641 if (msg->lm_buflens[segment] == newsize)
1644 /* nothing to do if we are enlarging the last segment */
1645 if (segment == msg->lm_bufcount - 1) {
1646 msg->lm_buflens[segment] = newsize;
1650 oldsize = msg->lm_buflens[segment];
1652 src = lustre_msg_buf(msg, segment + 1, 0);
1653 msg->lm_buflens[segment] = newsize;
1654 dst = lustre_msg_buf(msg, segment + 1, 0);
1655 msg->lm_buflens[segment] = oldsize;
1657 /* move from segment + 1 to end segment */
1658 LASSERT(msg->lm_magic == LUSTRE_MSG_MAGIC_V2);
1659 oldmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
1660 movesize = oldmsg_size - ((unsigned long) src - (unsigned long) msg);
1661 LASSERT(movesize >= 0);
1664 memmove(dst, src, movesize);
1666 /* note we don't clear the ares where old data live, not secret */
1668 /* finally set new segment size */
1669 msg->lm_buflens[segment] = newsize;
1671 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
1674 * Used by ptlrpc client to enlarge the \a segment of request message pointed
1675 * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
1676 * preserved after the enlargement. this must be called after original request
1677 * buffer being allocated.
1679 * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
1680 * so caller should refresh its local pointers if needed.
1682 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
1683 const struct req_msg_field *field,
1686 struct req_capsule *pill = &req->rq_pill;
1687 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1688 struct ptlrpc_sec_cops *cops;
1689 struct lustre_msg *msg = req->rq_reqmsg;
1690 int segment = __req_capsule_offset(pill, field, RCL_CLIENT);
1694 LASSERT(msg->lm_bufcount > segment);
1695 LASSERT(msg->lm_buflens[segment] <= newsize);
1697 if (msg->lm_buflens[segment] == newsize)
1700 cops = ctx->cc_sec->ps_policy->sp_cops;
1701 LASSERT(cops->enlarge_reqbuf);
1702 return cops->enlarge_reqbuf(ctx->cc_sec, req, segment, newsize);
1704 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
1707 * Used by ptlrpc client to allocate reply buffer of \a req.
1709 * \note After this, req->rq_repmsg is still not accessible.
1711 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
1713 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1714 struct ptlrpc_sec_policy *policy;
1719 LASSERT(ctx->cc_sec);
1720 LASSERT(ctx->cc_sec->ps_policy);
1725 policy = ctx->cc_sec->ps_policy;
1726 RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
1730 * Used by ptlrpc client to free reply buffer of \a req. After this
1731 * req->rq_repmsg is set to NULL and should not be accessed anymore.
1733 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
1735 struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
1736 struct ptlrpc_sec_policy *policy;
1741 LASSERT(ctx->cc_sec);
1742 LASSERT(ctx->cc_sec->ps_policy);
1743 LASSERT_ATOMIC_POS(&ctx->cc_refcount);
1745 if (req->rq_repbuf == NULL)
1747 LASSERT(req->rq_repbuf_len);
1749 policy = ctx->cc_sec->ps_policy;
1750 policy->sp_cops->free_repbuf(ctx->cc_sec, req);
1751 req->rq_repmsg = NULL;
1754 EXPORT_SYMBOL(sptlrpc_cli_free_repbuf);
1756 int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
1757 struct ptlrpc_cli_ctx *ctx)
1759 struct ptlrpc_sec_policy *policy = ctx->cc_sec->ps_policy;
1761 if (!policy->sp_cops->install_rctx)
1763 return policy->sp_cops->install_rctx(imp, ctx->cc_sec, ctx);
1766 int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
1767 struct ptlrpc_svc_ctx *ctx)
1769 struct ptlrpc_sec_policy *policy = ctx->sc_policy;
1771 if (!policy->sp_sops->install_rctx)
1773 return policy->sp_sops->install_rctx(imp, ctx);
1776 /* Get SELinux policy info from userspace */
1777 static int sepol_helper(struct obd_import *imp)
1779 char mtime_str[21] = { 0 }, mode_str[2] = { 0 };
1781 [0] = "/usr/sbin/l_getsepol",
1783 [2] = NULL, /* obd type */
1785 [4] = NULL, /* obd name */
1787 [6] = mtime_str, /* policy mtime */
1789 [8] = mode_str, /* enforcing mode */
1794 [1] = "PATH=/sbin:/usr/sbin",
1800 if (imp == NULL || imp->imp_obd == NULL ||
1801 imp->imp_obd->obd_type == NULL) {
1804 argv[2] = (char *)imp->imp_obd->obd_type->typ_name;
1805 argv[4] = imp->imp_obd->obd_name;
1806 spin_lock(&imp->imp_sec->ps_lock);
1807 if (imp->imp_sec->ps_sepol_mtime == 0 &&
1808 imp->imp_sec->ps_sepol[0] == '\0') {
1809 /* ps_sepol has not been initialized */
1813 snprintf(mtime_str, sizeof(mtime_str), "%lu",
1814 imp->imp_sec->ps_sepol_mtime);
1815 mode_str[0] = imp->imp_sec->ps_sepol[0];
1817 spin_unlock(&imp->imp_sec->ps_lock);
1818 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1825 static inline int sptlrpc_sepol_needs_check(struct ptlrpc_sec *imp_sec)
1829 if (send_sepol == 0 || !selinux_is_enabled())
1832 if (send_sepol == -1)
1833 /* send_sepol == -1 means fetch sepol status every time */
1836 spin_lock(&imp_sec->ps_lock);
1837 checknext = imp_sec->ps_sepol_checknext;
1838 spin_unlock(&imp_sec->ps_lock);
1840 /* next check is too far in time, please update */
1841 if (ktime_after(checknext,
1842 ktime_add(ktime_get(), ktime_set(send_sepol, 0))))
1845 if (ktime_before(ktime_get(), checknext))
1846 /* too early to fetch sepol status */
1850 /* define new sepol_checknext time */
1851 spin_lock(&imp_sec->ps_lock);
1852 imp_sec->ps_sepol_checknext = ktime_add(ktime_get(),
1853 ktime_set(send_sepol, 0));
1854 spin_unlock(&imp_sec->ps_lock);
1859 int sptlrpc_get_sepol(struct ptlrpc_request *req)
1861 struct ptlrpc_sec *imp_sec = req->rq_import->imp_sec;
1866 (req->rq_sepol)[0] = '\0';
1868 #ifndef HAVE_SELINUX
1869 if (unlikely(send_sepol != 0))
1871 "Client cannot report SELinux status, it was not built against libselinux.\n");
1875 if (send_sepol == 0 || !selinux_is_enabled())
1878 if (imp_sec == NULL)
1881 /* Retrieve SELinux status info */
1882 if (sptlrpc_sepol_needs_check(imp_sec))
1883 rc = sepol_helper(req->rq_import);
1884 if (likely(rc == 0)) {
1885 spin_lock(&imp_sec->ps_lock);
1886 memcpy(req->rq_sepol, imp_sec->ps_sepol,
1887 sizeof(req->rq_sepol));
1888 spin_unlock(&imp_sec->ps_lock);
1893 EXPORT_SYMBOL(sptlrpc_get_sepol);
1896 * server side security
1899 static int flavor_allowed(struct sptlrpc_flavor *exp,
1900 struct ptlrpc_request *req)
1902 struct sptlrpc_flavor *flvr = &req->rq_flvr;
1904 if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
1907 if ((req->rq_ctx_init || req->rq_ctx_fini) &&
1908 SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
1909 SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
1910 SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
1916 #define EXP_FLVR_UPDATE_EXPIRE (OBD_TIMEOUT_DEFAULT + 10)
1919 * Given an export \a exp, check whether the flavor of incoming \a req
1920 * is allowed by the export \a exp. Main logic is about taking care of
1921 * changing configurations. Return 0 means success.
1923 int sptlrpc_target_export_check(struct obd_export *exp,
1924 struct ptlrpc_request *req)
1926 struct sptlrpc_flavor flavor;
1932 * client side export has no imp_reverse, skip
1933 * FIXME maybe we should check flavor this as well???
1935 if (exp->exp_imp_reverse == NULL)
1938 /* don't care about ctx fini rpc */
1939 if (req->rq_ctx_fini)
1942 spin_lock(&exp->exp_lock);
1945 * if flavor just changed (exp->exp_flvr_changed != 0), we wait for
1946 * the first req with the new flavor, then treat it as current flavor,
1947 * adapt reverse sec according to it.
1948 * note the first rpc with new flavor might not be with root ctx, in
1949 * which case delay the sec_adapt by leaving exp_flvr_adapt == 1.
1951 if (unlikely(exp->exp_flvr_changed) &&
1952 flavor_allowed(&exp->exp_flvr_old[1], req)) {
1954 * make the new flavor as "current", and old ones as
1957 CDEBUG(D_SEC, "exp %p: just changed: %x->%x\n", exp,
1958 exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc);
1959 flavor = exp->exp_flvr_old[1];
1960 exp->exp_flvr_old[1] = exp->exp_flvr_old[0];
1961 exp->exp_flvr_expire[1] = exp->exp_flvr_expire[0];
1962 exp->exp_flvr_old[0] = exp->exp_flvr;
1963 exp->exp_flvr_expire[0] = ktime_get_real_seconds() +
1964 EXP_FLVR_UPDATE_EXPIRE;
1965 exp->exp_flvr = flavor;
1967 /* flavor change finished */
1968 exp->exp_flvr_changed = 0;
1969 LASSERT(exp->exp_flvr_adapt == 1);
1971 /* if it's gss, we only interested in root ctx init */
1972 if (req->rq_auth_gss &&
1973 !(req->rq_ctx_init &&
1974 (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
1975 req->rq_auth_usr_ost))) {
1976 spin_unlock(&exp->exp_lock);
1977 CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
1978 req->rq_auth_gss, req->rq_ctx_init,
1979 req->rq_auth_usr_root, req->rq_auth_usr_mdt,
1980 req->rq_auth_usr_ost);
1984 exp->exp_flvr_adapt = 0;
1985 spin_unlock(&exp->exp_lock);
1987 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
1988 req->rq_svc_ctx, &flavor);
1992 * if it equals to the current flavor, we accept it, but need to
1993 * dealing with reverse sec/ctx
1995 if (likely(flavor_allowed(&exp->exp_flvr, req))) {
1997 * most cases should return here, we only interested in
2000 if (!req->rq_auth_gss || !req->rq_ctx_init ||
2001 (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2002 !req->rq_auth_usr_ost)) {
2003 spin_unlock(&exp->exp_lock);
2008 * if flavor just changed, we should not proceed, just leave
2009 * it and current flavor will be discovered and replaced
2010 * shortly, and let _this_ rpc pass through
2012 if (exp->exp_flvr_changed) {
2013 LASSERT(exp->exp_flvr_adapt);
2014 spin_unlock(&exp->exp_lock);
2018 if (exp->exp_flvr_adapt) {
2019 exp->exp_flvr_adapt = 0;
2020 CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n",
2021 exp, exp->exp_flvr.sf_rpc,
2022 exp->exp_flvr_old[0].sf_rpc,
2023 exp->exp_flvr_old[1].sf_rpc);
2024 flavor = exp->exp_flvr;
2025 spin_unlock(&exp->exp_lock);
2027 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
2032 "exp %p (%x|%x|%x): is current flavor, install rvs ctx\n",
2033 exp, exp->exp_flvr.sf_rpc,
2034 exp->exp_flvr_old[0].sf_rpc,
2035 exp->exp_flvr_old[1].sf_rpc);
2036 spin_unlock(&exp->exp_lock);
2038 return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
2043 if (exp->exp_flvr_expire[0]) {
2044 if (exp->exp_flvr_expire[0] >= ktime_get_real_seconds()) {
2045 if (flavor_allowed(&exp->exp_flvr_old[0], req)) {
2047 "exp %p (%x|%x|%x): match the middle one (%lld)\n",
2048 exp, exp->exp_flvr.sf_rpc,
2049 exp->exp_flvr_old[0].sf_rpc,
2050 exp->exp_flvr_old[1].sf_rpc,
2051 (s64)(exp->exp_flvr_expire[0] -
2052 ktime_get_real_seconds()));
2053 spin_unlock(&exp->exp_lock);
2057 CDEBUG(D_SEC, "mark middle expired\n");
2058 exp->exp_flvr_expire[0] = 0;
2060 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match middle\n", exp,
2061 exp->exp_flvr.sf_rpc,
2062 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2063 req->rq_flvr.sf_rpc);
2067 * now it doesn't match the current flavor, the only chance we can
2068 * accept it is match the old flavors which is not expired.
2070 if (exp->exp_flvr_changed == 0 && exp->exp_flvr_expire[1]) {
2071 if (exp->exp_flvr_expire[1] >= ktime_get_real_seconds()) {
2072 if (flavor_allowed(&exp->exp_flvr_old[1], req)) {
2073 CDEBUG(D_SEC, "exp %p (%x|%x|%x): match the oldest one (%lld)\n",
2075 exp->exp_flvr.sf_rpc,
2076 exp->exp_flvr_old[0].sf_rpc,
2077 exp->exp_flvr_old[1].sf_rpc,
2078 (s64)(exp->exp_flvr_expire[1] -
2079 ktime_get_real_seconds()));
2080 spin_unlock(&exp->exp_lock);
2084 CDEBUG(D_SEC, "mark oldest expired\n");
2085 exp->exp_flvr_expire[1] = 0;
2087 CDEBUG(D_SEC, "exp %p (%x|%x|%x): %x not match found\n",
2088 exp, exp->exp_flvr.sf_rpc,
2089 exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_old[1].sf_rpc,
2090 req->rq_flvr.sf_rpc);
2092 CDEBUG(D_SEC, "exp %p (%x|%x|%x): skip the last one\n",
2093 exp, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc,
2094 exp->exp_flvr_old[1].sf_rpc);
2097 spin_unlock(&exp->exp_lock);
2099 CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with unauthorized flavor %x, expect %x|%x(%+lld)|%x(%+lld)\n",
2100 exp, exp->exp_obd->obd_name,
2101 req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
2102 req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
2103 req->rq_flvr.sf_rpc,
2104 exp->exp_flvr.sf_rpc,
2105 exp->exp_flvr_old[0].sf_rpc,
2106 exp->exp_flvr_expire[0] ?
2107 (s64)(exp->exp_flvr_expire[0] - ktime_get_real_seconds()) : 0,
2108 exp->exp_flvr_old[1].sf_rpc,
2109 exp->exp_flvr_expire[1] ?
2110 (s64)(exp->exp_flvr_expire[1] - ktime_get_real_seconds()) : 0);
2113 EXPORT_SYMBOL(sptlrpc_target_export_check);
2115 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
2116 struct sptlrpc_rule_set *rset)
2118 struct obd_export *exp;
2119 struct sptlrpc_flavor new_flvr;
2123 spin_lock(&obd->obd_dev_lock);
2125 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
2126 if (exp->exp_connection == NULL)
2130 * note if this export had just been updated flavor
2131 * (exp_flvr_changed == 1), this will override the
2134 spin_lock(&exp->exp_lock);
2135 sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
2136 exp->exp_connection->c_peer.nid,
2138 if (exp->exp_flvr_changed ||
2139 !flavor_equal(&new_flvr, &exp->exp_flvr)) {
2140 exp->exp_flvr_old[1] = new_flvr;
2141 exp->exp_flvr_expire[1] = 0;
2142 exp->exp_flvr_changed = 1;
2143 exp->exp_flvr_adapt = 1;
2145 CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
2146 exp, sptlrpc_part2name(exp->exp_sp_peer),
2147 exp->exp_flvr.sf_rpc,
2148 exp->exp_flvr_old[1].sf_rpc);
2150 spin_unlock(&exp->exp_lock);
2153 spin_unlock(&obd->obd_dev_lock);
2155 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
2157 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
2159 /* peer's claim is unreliable unless gss is being used */
2160 if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
2163 switch (req->rq_sp_from) {
2165 if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
2166 /* The below message is checked in sanity-sec test_33 */
2167 DEBUG_REQ(D_ERROR, req, "faked source CLI");
2168 svc_rc = SECSVC_DROP;
2172 if (!req->rq_auth_usr_mdt) {
2173 /* The below message is checked in sanity-sec test_33 */
2174 DEBUG_REQ(D_ERROR, req, "faked source MDT");
2175 svc_rc = SECSVC_DROP;
2179 if (!req->rq_auth_usr_ost) {
2180 /* The below message is checked in sanity-sec test_33 */
2181 DEBUG_REQ(D_ERROR, req, "faked source OST");
2182 svc_rc = SECSVC_DROP;
2187 if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
2188 !req->rq_auth_usr_ost) {
2189 /* The below message is checked in sanity-sec test_33 */
2190 DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
2191 svc_rc = SECSVC_DROP;
2196 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
2197 svc_rc = SECSVC_DROP;
2204 * Used by ptlrpc server, to perform transformation upon request message of
2205 * incoming \a req. This must be the first thing to do with an incoming
2206 * request in ptlrpc layer.
2208 * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
2209 * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
2210 * \retval SECSVC_COMPLETE success, the request has been fully processed, and
2211 * reply message has been prepared.
2212 * \retval SECSVC_DROP failed, this request should be dropped.
2214 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
2216 struct ptlrpc_sec_policy *policy;
2217 struct lustre_msg *msg = req->rq_reqbuf;
2223 LASSERT(req->rq_reqmsg == NULL);
2224 LASSERT(req->rq_repmsg == NULL);
2225 LASSERT(req->rq_svc_ctx == NULL);
2227 req->rq_req_swab_mask = 0;
2229 rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
2232 lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
2236 CERROR("error unpacking request from %s x%llu\n",
2237 libcfs_id2str(req->rq_peer), req->rq_xid);
2238 RETURN(SECSVC_DROP);
2241 req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
2242 req->rq_sp_from = LUSTRE_SP_ANY;
2243 req->rq_auth_uid = -1; /* set to INVALID_UID */
2244 req->rq_auth_mapped_uid = -1;
2246 policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
2248 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
2249 RETURN(SECSVC_DROP);
2252 LASSERT(policy->sp_sops->accept);
2253 rc = policy->sp_sops->accept(req);
2254 sptlrpc_policy_put(policy);
2255 LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
2256 LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
2259 * if it's not null flavor (which means embedded packing msg),
2260 * reset the swab mask for the comming inner msg unpacking.
2262 if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
2263 req->rq_req_swab_mask = 0;
2265 /* sanity check for the request source */
2266 rc = sptlrpc_svc_check_from(req, rc);
2271 * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
2272 * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
2273 * a buffer of \a msglen size.
2275 int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
2277 struct ptlrpc_sec_policy *policy;
2278 struct ptlrpc_reply_state *rs;
2283 LASSERT(req->rq_svc_ctx);
2284 LASSERT(req->rq_svc_ctx->sc_policy);
2286 policy = req->rq_svc_ctx->sc_policy;
2287 LASSERT(policy->sp_sops->alloc_rs);
2289 rc = policy->sp_sops->alloc_rs(req, msglen);
2290 if (unlikely(rc == -ENOMEM)) {
2291 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
2293 if (svcpt->scp_service->srv_max_reply_size <
2294 msglen + sizeof(struct ptlrpc_reply_state)) {
2295 /* Just return failure if the size is too big */
2296 CERROR("size of message is too big (%zd), %d allowed\n",
2297 msglen + sizeof(struct ptlrpc_reply_state),
2298 svcpt->scp_service->srv_max_reply_size);
2302 /* failed alloc, try emergency pool */
2303 rs = lustre_get_emerg_rs(svcpt);
2307 req->rq_reply_state = rs;
2308 rc = policy->sp_sops->alloc_rs(req, msglen);
2310 lustre_put_emerg_rs(rs);
2311 req->rq_reply_state = NULL;
2316 (req->rq_reply_state && req->rq_reply_state->rs_msg));
2322 * Used by ptlrpc server, to perform transformation upon reply message.
2324 * \post req->rq_reply_off is set to approriate server-controlled reply offset.
2325 * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
2327 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
2329 struct ptlrpc_sec_policy *policy;
2334 LASSERT(req->rq_svc_ctx);
2335 LASSERT(req->rq_svc_ctx->sc_policy);
2337 policy = req->rq_svc_ctx->sc_policy;
2338 LASSERT(policy->sp_sops->authorize);
2340 rc = policy->sp_sops->authorize(req);
2341 LASSERT(rc || req->rq_reply_state->rs_repdata_len);
2347 * Used by ptlrpc server, to free reply_state.
2349 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
2351 struct ptlrpc_sec_policy *policy;
2352 unsigned int prealloc;
2356 LASSERT(rs->rs_svc_ctx);
2357 LASSERT(rs->rs_svc_ctx->sc_policy);
2359 policy = rs->rs_svc_ctx->sc_policy;
2360 LASSERT(policy->sp_sops->free_rs);
2362 prealloc = rs->rs_prealloc;
2363 policy->sp_sops->free_rs(rs);
2366 lustre_put_emerg_rs(rs);
2370 void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
2372 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2375 atomic_inc(&ctx->sc_refcount);
2378 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
2380 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2385 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2386 if (atomic_dec_and_test(&ctx->sc_refcount)) {
2387 if (ctx->sc_policy->sp_sops->free_ctx)
2388 ctx->sc_policy->sp_sops->free_ctx(ctx);
2390 req->rq_svc_ctx = NULL;
2393 void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
2395 struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
2400 LASSERT_ATOMIC_POS(&ctx->sc_refcount);
2401 if (ctx->sc_policy->sp_sops->invalidate_ctx)
2402 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
2404 EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
2411 * Perform transformation upon bulk data pointed by \a desc. This is called
2412 * before transforming the request message.
2414 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
2415 struct ptlrpc_bulk_desc *desc)
2417 struct ptlrpc_cli_ctx *ctx;
2419 LASSERT(req->rq_bulk_read || req->rq_bulk_write);
2421 if (!req->rq_pack_bulk)
2424 ctx = req->rq_cli_ctx;
2425 if (ctx->cc_ops->wrap_bulk)
2426 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
2429 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
2432 * This is called after unwrap the reply message.
2433 * return nob of actual plain text size received, or error code.
2435 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
2436 struct ptlrpc_bulk_desc *desc,
2439 struct ptlrpc_cli_ctx *ctx;
2442 LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
2444 if (!req->rq_pack_bulk)
2445 return desc->bd_nob_transferred;
2447 ctx = req->rq_cli_ctx;
2448 if (ctx->cc_ops->unwrap_bulk) {
2449 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2453 return desc->bd_nob_transferred;
2455 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
2458 * This is called after unwrap the reply message.
2459 * return 0 for success or error code.
2461 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
2462 struct ptlrpc_bulk_desc *desc)
2464 struct ptlrpc_cli_ctx *ctx;
2467 LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
2469 if (!req->rq_pack_bulk)
2472 ctx = req->rq_cli_ctx;
2473 if (ctx->cc_ops->unwrap_bulk) {
2474 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
2480 * if everything is going right, nob should equals to nob_transferred.
2481 * in case of privacy mode, nob_transferred needs to be adjusted.
2483 if (desc->bd_nob != desc->bd_nob_transferred) {
2484 CERROR("nob %d doesn't match transferred nob %d\n",
2485 desc->bd_nob, desc->bd_nob_transferred);
2491 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
2493 #ifdef HAVE_SERVER_SUPPORT
2495 * Performe transformation upon outgoing bulk read.
2497 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
2498 struct ptlrpc_bulk_desc *desc)
2500 struct ptlrpc_svc_ctx *ctx;
2502 LASSERT(req->rq_bulk_read);
2504 if (!req->rq_pack_bulk)
2507 ctx = req->rq_svc_ctx;
2508 if (ctx->sc_policy->sp_sops->wrap_bulk)
2509 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
2513 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
2516 * Performe transformation upon incoming bulk write.
2518 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
2519 struct ptlrpc_bulk_desc *desc)
2521 struct ptlrpc_svc_ctx *ctx;
2524 LASSERT(req->rq_bulk_write);
2527 * if it's in privacy mode, transferred should >= expected; otherwise
2528 * transferred should == expected.
2530 if (desc->bd_nob_transferred < desc->bd_nob ||
2531 (desc->bd_nob_transferred > desc->bd_nob &&
2532 SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
2533 SPTLRPC_BULK_SVC_PRIV)) {
2534 DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
2535 desc->bd_nob_transferred, desc->bd_nob);
2539 if (!req->rq_pack_bulk)
2542 ctx = req->rq_svc_ctx;
2543 if (ctx->sc_policy->sp_sops->unwrap_bulk) {
2544 rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
2546 CERROR("error unwrap bulk: %d\n", rc);
2549 /* return 0 to allow reply be sent */
2552 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
2555 * Prepare buffers for incoming bulk write.
2557 int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
2558 struct ptlrpc_bulk_desc *desc)
2560 struct ptlrpc_svc_ctx *ctx;
2562 LASSERT(req->rq_bulk_write);
2564 if (!req->rq_pack_bulk)
2567 ctx = req->rq_svc_ctx;
2568 if (ctx->sc_policy->sp_sops->prep_bulk)
2569 return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
2573 EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
2575 #endif /* HAVE_SERVER_SUPPORT */
2578 * user descriptor helpers
2581 int sptlrpc_current_user_desc_size(void)
2585 ngroups = current_ngroups;
2587 if (ngroups > LUSTRE_MAX_GROUPS)
2588 ngroups = LUSTRE_MAX_GROUPS;
2589 return sptlrpc_user_desc_size(ngroups);
2591 EXPORT_SYMBOL(sptlrpc_current_user_desc_size);
2593 int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
2595 struct ptlrpc_user_desc *pud;
2597 pud = lustre_msg_buf(msg, offset, 0);
2599 pud->pud_uid = from_kuid(&init_user_ns, current_uid());
2600 pud->pud_gid = from_kgid(&init_user_ns, current_gid());
2601 pud->pud_fsuid = from_kuid(&init_user_ns, current_fsuid());
2602 pud->pud_fsgid = from_kgid(&init_user_ns, current_fsgid());
2603 pud->pud_cap = cfs_curproc_cap_pack();
2604 pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
2607 if (pud->pud_ngroups > current_ngroups)
2608 pud->pud_ngroups = current_ngroups;
2609 #ifdef HAVE_GROUP_INFO_GID
2610 memcpy(pud->pud_groups, current_cred()->group_info->gid,
2611 pud->pud_ngroups * sizeof(__u32));
2612 #else /* !HAVE_GROUP_INFO_GID */
2613 memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
2614 pud->pud_ngroups * sizeof(__u32));
2615 #endif /* HAVE_GROUP_INFO_GID */
2616 task_unlock(current);
2620 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
2622 int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
2624 struct ptlrpc_user_desc *pud;
2627 pud = lustre_msg_buf(msg, offset, sizeof(*pud));
2632 __swab32s(&pud->pud_uid);
2633 __swab32s(&pud->pud_gid);
2634 __swab32s(&pud->pud_fsuid);
2635 __swab32s(&pud->pud_fsgid);
2636 __swab32s(&pud->pud_cap);
2637 __swab32s(&pud->pud_ngroups);
2640 if (pud->pud_ngroups > LUSTRE_MAX_GROUPS) {
2641 CERROR("%u groups is too large\n", pud->pud_ngroups);
2645 if (sizeof(*pud) + pud->pud_ngroups * sizeof(__u32) >
2646 msg->lm_buflens[offset]) {
2647 CERROR("%u groups are claimed but bufsize only %u\n",
2648 pud->pud_ngroups, msg->lm_buflens[offset]);
2653 for (i = 0; i < pud->pud_ngroups; i++)
2654 __swab32s(&pud->pud_groups[i]);
2659 EXPORT_SYMBOL(sptlrpc_unpack_user_desc);
2665 const char *sec2target_str(struct ptlrpc_sec *sec)
2667 if (!sec || !sec->ps_import || !sec->ps_import->imp_obd)
2669 if (sec_is_reverse(sec))
2671 return obd_uuid2str(&sec->ps_import->imp_obd->u.cli.cl_target_uuid);
2673 EXPORT_SYMBOL(sec2target_str);
2676 * return true if the bulk data is protected
2678 int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
2680 switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
2681 case SPTLRPC_BULK_SVC_INTG:
2682 case SPTLRPC_BULK_SVC_PRIV:
2688 EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
2691 * crypto API helper/alloc blkciper
2695 * initialize/finalize
2698 int sptlrpc_init(void)
2702 rwlock_init(&policy_lock);
2704 rc = sptlrpc_gc_init();
2708 rc = sptlrpc_conf_init();
2712 rc = sptlrpc_enc_pool_init();
2716 rc = sptlrpc_null_init();
2720 rc = sptlrpc_plain_init();
2724 rc = sptlrpc_lproc_init();
2731 sptlrpc_plain_fini();
2733 sptlrpc_null_fini();
2735 sptlrpc_enc_pool_fini();
2737 sptlrpc_conf_fini();
2744 void sptlrpc_fini(void)
2746 sptlrpc_lproc_fini();
2747 sptlrpc_plain_fini();
2748 sptlrpc_null_fini();
2749 sptlrpc_enc_pool_fini();
2750 sptlrpc_conf_fini();